There's Personalized Medicine. Why Not Personalized Inference?

• Activity Number: 339
• Type: Invited
• Date/Time: Tuesday, August 11, 2015 : 10:30 AM to 12:20 PM
• Sponsor: International Society for Bayesian Analysis (ISBA)
• Author(s): Xiao-Li Meng* and Keli Liu
• Companies: Harvard University and Stanford University
• Keywords: Resolution ; Robustness ; Relevance ; Bayesian ; Frequentist ; Fiducial

Abstract

Interest in personalized treatments, from medicine to education to marketing, has surged in recent years. It arises from the recognition that a treatment with 95% effectiveness over the population may exhibit a much lower success rate for "me". In contrast, the trend in statistics has been to develop "off-the shelf" procedures which require no personalization to the problem at hand. In simple situations, convenience rightly takes precedence: why go to the doctor when off the counter medication suffices to treat a sore throat? But with increasingly complex data in this era of Big Data, one wonders, "How applicable is the promised 95% confidence of an off-the-shelf confidence interval in assessing its performance on my dataset?". Just as doctors personalize treatment, statisticians need to reorient themselves to delivering personalized inferences. Identifying the appropriate degree of personalization for a particular problem is a key challenge in statistical inference. This personalized (re)-view is inevitably idiosyncratic, but we hope the reader enjoys our renovations on this tour of the century old world of inference.